Antidepressants differ by 4 kg and 21 beats per minute, network analysis finds

Antidepressants are usually compared on whether they work and whether people stay on them. A network meta-analysis published in The Lancet on October 21 asks a different question — what they do to the body — and finds that the drugs are much less interchangeable on that axis than on efficacy [s1].

What was pooled

The authors searched MEDLINE, EMBASE, PsycINFO, ClinicalTrials.gov and the US Food and Drug Administration website from inception to April 21, 2025 [s1]. They included single- and double-blinded randomised controlled trials comparing antidepressants with placebo in acute monotherapy for any psychiatric disorder [s1].

Of 26,252 citations screened, 151 studies and 17 FDA reports met the criteria, covering 58,534 participants and comparing 30 antidepressants with placebo [s1]. Median treatment duration was eight weeks (IQR 6.0-8.5) [s1].

The outcomes were physiological, not psychiatric: weight; total cholesterol; glucose; heart rate; systolic and diastolic blood pressure; corrected QT interval; sodium; potassium; aspartate transferase; alanine transaminase; alkaline phosphatase; bilirubin; urea; and creatinine [s1]. Frequentist random-effects network meta-analysis was used, with meta-regressions examining associations with age, sex and baseline weight [s1].

The size of the differences

The spread between drugs is the finding.

Weight change differed by approximately 4 kg between agomelatine and maprotiline [s1]. Heart rate change differed by over 21 beats per minute between fluvoxamine and nortriptyline [s1]. Systolic blood pressure differed by over 11 mmHg between nortriptyline and doxepin [s1].

These are between-drug ranges within a single class of medicines prescribed for the same indications, over a median of eight weeks.

The dissociation worth noticing

Paroxetine, duloxetine, desvenlafaxine and venlafaxine were associated with increases in total cholesterol, and duloxetine additionally with increases in glucose — despite all of these drugs reducing bodyweight [s1].

That decouples two things routinely bundled together. The assumption that a drug which does not cause weight gain is metabolically neutral does not hold across this dataset.

There was also strong evidence that duloxetine, desvenlafaxine and levomilnacipran increased AST, ALT and ALP concentrations, although the magnitudes were not considered clinically significant [s1].

What the analysis did not find

The authors did not find strong evidence that any antidepressant affected the corrected QT interval, or concentrations of sodium, potassium, urea or creatinine, to a clinically significant extent [s1].

QTc in particular is worth flagging, because it drives a substantial amount of monitoring practice. This analysis, over a median of eight weeks in trial populations, did not detect clinically significant effects [s1].

Who is most affected

Two meta-regression findings point to differential risk. Higher baseline bodyweight was associated with larger antidepressant-induced increases in systolic blood pressure, ALT and AST [s1]. Higher baseline age was associated with larger antidepressant-induced increases in glucose [s1].

The authors also found no association between changes in depressive symptoms and metabolic disturbance [s1]. The physiological effects are not a by-product of getting better or failing to.

What the design constrains

Median treatment duration was eight weeks [s1]. Antidepressants are frequently taken for years. An eight-week weight or cholesterol trajectory is a starting gradient, not an endpoint, and there is no basis here for extrapolating either direction — effects could accumulate or plateau.

These are randomised trial populations, which are systematically healthier than the people who take these drugs in practice: fewer comorbidities, less polypharmacy, more monitoring. That cuts against generalising the absolute magnitudes.

Network meta-analysis estimates comparisons between drugs that were mostly never tested against each other, by routing through their shared placebo comparisons. The method is well established and the assumption it rests on — that trials of different drugs are similar enough to be connected — is the standard place these analyses can go wrong. With 30 drugs and 151 studies [s1], some comparisons will rest on much thinner evidence than others.

Where it sits against the efficacy literature

The best-known network meta-analysis of antidepressants, published in the same journal in 2018, compared efficacy and acceptability across 21 drugs for the acute treatment of adults with major depressive disorder [s2]. It has shaped prescribing guidance since.

The October analysis is the physiological counterpart to that work: same method, same journal, different outcome set. Its authors argue that treatment guidelines should be updated to reflect differences in physiological risk, while stating that the choice of antidepressant should be made on an individual basis, considering clinical presentation and the preferences of patients, carers and clinicians [s1].

That second clause is not throat-clearing. A drug that raises cholesterol may still be the right drug for someone it helps when others have not.

What to watch

Whether guideline bodies incorporate physiological ranking alongside efficacy ranking; whether longer-term observational data show these eight-week gradients persisting; and whether the cholesterol-without-weight finding for the SNRIs is replicated in cohorts followed for years rather than weeks.

This article describes pooled trial evidence and is informational only. It is not medical advice, and nothing here should be used to start, stop or change any medication.

Sources

• [s1] The effects of antidepressants on cardiometabolic and other physiological parameters: a systematic review and network meta-analysis, The Lancet, 2025;406(10515):2063-2077, published online 2025-10-21.
• [s2] Comparative efficacy and acceptability of 21 antidepressant drugs for the acute treatment of adults with major depressive disorder: a systematic review and network meta-analysis, The Lancet, 2018;391(10128):1357-1366, published online 2018-02-21.